Verification of Geometric Robustness of Neural Networks via Piecewise Linear Approximation and Lipschitz Optimisation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Batten, Ben, Zheng, Yang, De Palma, Alessandro, Kouvaros, Panagiotis, Lomuscio, Alessio
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910615627366400
author Batten, Ben
Zheng, Yang
De Palma, Alessandro
Kouvaros, Panagiotis
Lomuscio, Alessio
author_facet Batten, Ben
Zheng, Yang
De Palma, Alessandro
Kouvaros, Panagiotis
Lomuscio, Alessio
contents We address the problem of verifying neural networks against geometric transformations of the input image, including rotation, scaling, shearing, and translation. The proposed method computes provably sound piecewise linear constraints for the pixel values by using sampling and linear approximations in combination with branch-and-bound Lipschitz optimisation. The method obtains provably tighter over-approximations of the perturbation region than the present state-of-the-art. We report results from experiments on a comprehensive set of verification benchmarks on MNIST and CIFAR10. We show that our proposed implementation resolves up to 32% more verification cases than present approaches.
format Preprint
id arxiv_https___arxiv_org_abs_2408_13140
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Verification of Geometric Robustness of Neural Networks via Piecewise Linear Approximation and Lipschitz Optimisation
Batten, Ben
Zheng, Yang
De Palma, Alessandro
Kouvaros, Panagiotis
Lomuscio, Alessio
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
We address the problem of verifying neural networks against geometric transformations of the input image, including rotation, scaling, shearing, and translation. The proposed method computes provably sound piecewise linear constraints for the pixel values by using sampling and linear approximations in combination with branch-and-bound Lipschitz optimisation. The method obtains provably tighter over-approximations of the perturbation region than the present state-of-the-art. We report results from experiments on a comprehensive set of verification benchmarks on MNIST and CIFAR10. We show that our proposed implementation resolves up to 32% more verification cases than present approaches.
title Verification of Geometric Robustness of Neural Networks via Piecewise Linear Approximation and Lipschitz Optimisation
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2408.13140